> “Undeniable proof that confidently uninformed hospital administrators are a danger to patients: easily duped by AI companies that are nowhere near capable of providing patient care,” [Radiologist Dr.] Suhail told Radiology Business. “Any attempt to implement AI-only reads would immediately result in patient harm and death, and only someone with zero understanding of radiology would say something so naive. But in some sense, they’re correct: Hospitals are happy to cut costs even if it means patient harm, as long as it’s legal.”

[1] https://karlbode.com/ceo-said-a-thing-journalism/

The public has lost confidence in science because commercial and political entities have worked systematically to undermine scientific authority that threatens their business models and narratives.

Fossil fuels, tobacco, and refined sugar single-handedly manufactured decades of scientific "controversy" around climate change, tobacco's health impact, and the role of sugars and fats in obesity and heart disease. Religious fundamentalists fund pseudo-scientific books and articles attempting to muddy the waters concerning geological and biological evidence about any time period before the invention of agriculture. Grifters fabricate arguments and data to delegitimize Western medicine to sell placebos at high markup. Politicians attribute all unflattering research to partisan skullduggery.

It is true that hard science, social science, and medicine all have published flawed work; sometimes even in bad faith. These occasional failures -- which should be corrected and never tolerated -- are then used by the same people to indict the entire enterprise.

It's true that there's research misconduct, and it must be weeded out.

What's also true is that funding for research is declining in the US, especially relative to our need for it.

Note how many "tweaks" center on things like sample-size or statistical significance. Human-oriented research is always intractably complex, and we get the scientific outcomes we do because studies are perennially understaffed and under-resourced relative to the questions they seek to answer. Trying to answer fundamental questions about health and wellness with 30 people tracked over 6 weeks is fundamentally flawed, but it's the best we can do with the resources we have.

In a world where funding was plentiful, and career paths not cut-throat and perilous, there'd be far fewer examples of these kinds of "tweaks." People respond to incentives, and if the options are "add a few more participants, because after all the initial sample size was somewhat arbitrary" or "fail to publish, fail to graduate, fail to get a permanent position after a decade of post-secondary education because applicants must be perfect," the only surprise is that they aren't more common.

The link you sent basically says "you pinky-promise that you're 18."

Later in the doc it also says "you promise not to send malware."

On Russian, the explanations of why some answers were correct/incorrect didn't load (presumably an AI call failed?). Especially at lower layers, a good fallback would be a simple dictionary definition.

On Spanish, I did the placement test, then it asked which "dialect" I wanted. I selected Mexican, and was treated with truly excellent renderings of European pronunciation. I wouldn't have been mad if all it had was one set of pronunciation, and it's more frustrating to see the ignored option than to never have it at all.

As for the placement test: I got dropped into lesson 2 for Spanish. For comparison, I placed into Lesson 5 in Russian, where I actually got more incorrect answers. The Spanish placement test wasn't very deep, and I knew all the answers. It told me I got two wrong, so either the test is wrong or I just got punchy and hit the wrong buttons.

Recommendation: scale back on the ambition. Focus on getting the educational and product experience right with languages you know first. Be honest about data provenance and limitations.

But the things that irritate me even more are the infernal modals and alerts on my computing devices. It is hard enough maintaining focus without having to spend an entire work session playing whack-a-mole at random intervals for a hundred different things that aren’t relevant. I never want to know that my scanner software has an update available.

I realized that at its core, this problem is caused by developers and product managers mistakenly believing that I care as much about their product as they do.

It would be nice if the gatekeepers had mechanisms that punished this behavior. Search engines should lower the rankings of every site with random modals. App stores could display a normalized metric of alert click through — “this app has an above average number of alerts that are ignored”.

This article is fiction followed by commentary.

If the anecdote were real, it would still be an anecdote: tempting to generalize, but wise to hold loosely.

The article is about how AI will lead to a labor surplus in certain professions, while other professions will retain employment.

The article compares this to the Black Death, where labor supply decreased uniformly. Labor was then able to extract concessions from capital.

Industrialization leads to a different outcome: capital captures more value initially, while devastating workers in the short-term. In the long-term, everyone benefits from higher living standards. Even if the article is correct about AI impacts, it doesn't explain why AI is different from all prior industrialization.

Your comment about “1700’s newspapers” reminded me of The Past Times podcast, where comedians read random newspapers from across American history. The episodes I’ve listened to were delightful, and they covered mundane news in mundane places.

“O brother, where art thou” is one of my favorite movies. It’s a retelling of The Odyssey (a literally prehistoric tale) set in Depression-era Mississippi, made in the early 2000’s.

The specific question of editing out these production artifacts doesn’t rile me either way, though. I didn’t see the original mistake, and I won’t notice the fix either.

I’ll also agree that just as no one steps in the same river twice, how the past is viewed and interpreted changes over time. What is valued or not also changes. 90% of everything is still crap. And quite a bit of the interest in the past is reflected in remixes or retellings for modern audiences.

Still, people also read Beowulf or Chaucer in the original or in modern translation. Others will enjoy both Jane Austen and Bridgerton. People will listen to Beethoven and Jon Batiste. Sure, not all those things are for everyone, but neither are modern music genres, sports entertainment, or most TV shows.

Any discussion of public media trust that doesn’t include this as a component of analysis is immediately suspect.

An analysis that claims that public mistrust in media is because the media did not create more space for right wing obfuscation and disinformation is at best misguided. In practice, it is more likely another element of the campaign that created the problem in the first place.

There are many valid criticisms of the modern press. “They didn’t conform to the right wing’s warped presentation of reality” is not one of them.

I agree root cause analysis would be more interesting, but it wouldn't be justified until the base phenomena was validated.

Sure, people who do exercise think it helps stress and anxiety, but lots of people also find homeopathic remedies to be helpful. Papers like this show the former stand up under experimentation and the later don't.

Why isn’t this a serious problem then? I’d love a networking expert to chime in.

Is it that high bandwidth links also have very low packet error rates?

Or is it that individual TCP flows rarely saturate the link? (Because of congestion control, lower end to end throughput, sharing links, or some other reason?)

The problem is so bad in the discipline they changed how they name things: https://www.theverge.com/2020/8/6/21355674/human-genes-renam...

OP’s post is probably still a good idea for researchers, where something like SPSS offers better protection for datasets, albeit with a higher learning curve.

For the rest of us, so long as we pay attention, Excel is an incredibly underrated analysis approach.

TFA states the facility will use gray water, but I think it’s good of watchdogs to make sure that stays the case.

I don’t see how removing a guaranteed market will spur investment in the capital intensive activity of ship building.

It also occurs to me to ask how much domestic shipping traffic there is in the first place, and if it is actually price-sensitive. Unlike Japan and Korea, most of the US is far from coastal and riverine shipping lanes.

Chinese students who want to study this topic in the US face extreme scrutiny in visa interviews. On cursory inspection, there is a logic there.

The logic falls apart when you realize that all academic research is intended to be published. We don’t develop secret IP or trade secrets[1]. Our development offices want us to obtain patents on our work, but that is another form of publication.

I wish the powers that are moving to stigmatize Chinese academics would be clearer why they are doing it.

If they are concerned that students are using a student visa as a cover story for traditional espionage (cultivating assets, etc) they should say so.

If they are doing it in an attempt to starve China of trained workers in cutting edge tech, like they are with chip bans, I wish they would say so.

But claiming this is about “IP theft” makes it unclear if these efforts are pretextual or if our national security apparatus legitimately doesn’t understand the basics of how universities work.

[1] Some labs do classified work, but are limited to cleared citizens.

Google Fiber already offers 2Gbps to a friend of mine, but he hasn't upgraded because he doesn't see the point. Most of his bulk transfers (Steam, etc.) seem to be throttled at the source, and upgrading from Gigabit to 2.5GbE or 10GbE gear isn't cheap.

I could see this being useful for an office building or an apartment building, but what would a single family home do with it?

At my (public) institution, none of these identities were discussed when considering candidates.

We did look at diversity statements as an indicator of if applicants were aware of the broadening participation requirements set by federal funding agencies.

As an aside, the experiences of the marginalized people you describe don't seem to indicate that there is a net advantage to them. I think anyone who was adopting an identity in bad faith would quickly discover that it wasn't as helpful as they thought.

TP-Link has been good about deploying improvements to the ecosystem.

The flipside is those improvements are basic functionality that should have been present in a commercial product from v1. Like the ability to set firewall rules, which came in late 2021 or 22 iirc.

Other issues:

- The Safari interface issue she mentions is a major pain, and it’s a regression.

- All of their equipment has multiple “versions” under the same model number, and the early adopters like me get updates last for some reason.

- Documentation is ok, but logging and troubleshooting is difficult with the limited interface you get.

- DHCP IP reservation was broken until the last release.

I am glad the ecosystem is improving, and I appreciate having visibility into the network from a single pane of glass.

For a home network I can tolerate these issues. But I would be furious if I were trying to run a business network with it.